You are viewing documentation for Kubernetes version: v1.28

Kubernetes v1.28 documentation is no longer actively maintained. The version you are currently viewing is a static snapshot. For up-to-date information, see the latest version.

kubeadm: Use etcd Learner to Join a Control Plane Node Safely

Author: Paco Xu (DaoCloud)

The kubeadm tool now supports etcd learner mode, which allows you to enhance the resilience and stability of your Kubernetes clusters by leveraging the learner mode feature introduced in etcd version 3.4. This guide will walk you through using etcd learner mode with kubeadm. By default, kubeadm runs a local etcd instance on each control plane node.

In v1.27, kubeadm introduced a new feature gate EtcdLearnerMode. With this feature gate enabled, when joining a new control plane node, a new etcd member will be created as a learner and promoted to a voting member only after the etcd data are fully aligned.

What are the advantages of using etcd learner mode?

etcd learner mode offers several compelling reasons to consider its adoption in Kubernetes clusters:

  1. Enhanced Resilience: etcd learner nodes are non-voting members that catch up with the leader's logs before becoming fully operational. This prevents new cluster members from disrupting the quorum or causing leader elections, making the cluster more resilient during membership changes.
  2. Reduced Cluster Unavailability: Traditional approaches to adding new members often result in cluster unavailability periods, especially in slow infrastructure or misconfigurations. etcd learner mode minimizes such disruptions.
  3. Simplified Maintenance: Learner nodes provide a safer and reversible way to add or replace cluster members. This reduces the risk of accidental cluster outages due to misconfigurations or missteps during member additions.
  4. Improved Network Tolerance: In scenarios involving network partitions, learner mode allows for more graceful handling. Depending on the partition a new member lands, it can seamlessly integrate with the existing cluster without causing disruptions.

In summary, the etcd learner mode improves the reliability and manageability of Kubernetes clusters during member additions and changes, making it a valuable feature for cluster operators.

How nodes join a cluster that's using the new mode

Create a Kubernetes cluster backed by etcd in learner mode

For a general explanation about creating highly available clusters with kubeadm, you can refer to Creating Highly Available Clusters with kubeadm.

To create a Kubernetes cluster, backed by etcd in learner mode, using kubeadm, follow these steps:

# kubeadm init --feature-gates=EtcdLearnerMode=true ...
kubeadm init --config=kubeadm-config.yaml

The kubeadm configuration file is like below:

kind: ClusterConfiguration
  EtcdLearnerMode: true

The kubeadm tool deploys a single-node Kubernetes cluster with etcd set to use learner mode.

Join nodes to the Kubernetes cluster

Before joining a control-plane node to the new Kubernetes cluster, ensure that the existing control plane nodes and all etcd members are healthy.

Check the cluster health with etcdctl. If etcdctl isn't available, you can run this tool inside a container image. You would do that directly with your container runtime using a tool such as crictl run and not through Kubernetes

Here is an example on a client command that uses secure communication to check the cluster health of the etcd cluster:

ETCDCTL_API=3 etcdctl --endpoints \
  --cert=/etc/kubernetes/pki/etcd/server.crt \
  --key=/etc/kubernetes/pki/etcd/server.key \
  --cacert=/etc/kubernetes/pki/etcd/ca.crt \
  member list
dc543c4d307fadb9, started, node1,,, false

To check if the Kubernetes control plane is healthy, run kubectl get node -l and check if the nodes are ready.

Before joining a worker node to the new Kubernetes cluster, ensure that the control plane nodes are healthy.

What's next


Was this guide helpful? If you have any feedback or encounter any issues, please let us know. Your feedback is always welcome! Join the bi-weekly SIG Cluster Lifecycle meeting or weekly kubeadm office hours. Or reach us via Slack (channel #kubeadm), or the SIG's mailing list.